Prediction of accrual expenses in balance sheet using decision trees and linear regression

Abstract

In response to globalization, International Financial Reporting Standards (IFRS) has become the norm of the global capital markets. Companies preparing financial statements using IFRS may make the financial situation fully disclosed. Nevertheless, an overestimated accrual expense of a balance sheet may not only underestimate the earnings data, but also increase the cash outflows of the statement of cash flows. When the accrual expense is underestimated, corporate earnings will inflate earnings statistics. In addition, the problem of funds shortage may occur upon actual payment because the cash outflows of the statement of cash flows is underestimated. In this paper, we adopt the prediction mechanism in data mining to predict the unused vacation time of employees, which in turn becomes a part of the accrual expenses in the balance sheet. The prediction target is the bonus of unused annual leave in terms of unused hours so that the estimated amount of fees payable accuracy in the balance sheet can be improved. Both decision-tree models and regression analysis are used. Comprehensive experiments show that the decision-tree method outperforms the regression analysis method, with MAE of −23.1 and RMSE of 43.1.